DeepSeek raises V4 API prices by up to 1,100% just as Chinese AI startup launches DeepSeek-V4-Pro

DeepSeek is making its newest AI model more expensive almost as soon as it arrives.

The Chinese AI startup said Thursday it will raise API prices for its V4-Pro and V4-Flash models, with increases ranging from 50% to 1,100% depending on the model, token type, and time of use. The new rates take effect August 17 and introduce peak and off-peak pricing, marking a significant shift in DeepSeek’s aggressive low-cost strategy.

“The new rates, ranging from 50% to 1,100% above current prices depending on the model, token type, and time of use, will take effect on August 17,” Reuters reported, citing a company statement.

The timing is hard to miss. DeepSeek announced the pricing changes on the same day it launched DeepSeek-V4-Pro, its latest model aimed at coding agents and complex autonomous tasks.

Under the hood, V4 Pro uses a 1.6 trillion-parameter mixture-of-experts architecture and supports a 1 million-token context window. DeepSeek initially priced the model at $0.435 per million input tokens and $0.87 per million output tokens, keeping with the company’s reputation for putting pressure on rivals through unusually low AI inference costs.

That pricing equation is now changing.

DeepSeek Launches New V4 Pro Model to Challenge Claude

The move to peak and off-peak pricing suggests DeepSeek is beginning to confront one of the harder economics of operating frontier AI models at scale: demand is not evenly distributed.

Variable pricing could give developers an incentive to move less time-sensitive workloads into cheaper periods, helping DeepSeek manage infrastructure demand. For companies running large numbers of agent tasks, though, increases reaching 1,100% could materially change the economics that made DeepSeek attractive in the first place.

That creates an interesting test for the company. DeepSeek built much of its global reputation around a simple proposition: advanced AI performance did not have to carry Silicon Valley-level prices. Developers could access increasingly capable models at a fraction of what competing frontier systems cost.

V4 Pro pushes the performance side of that argument further. The new pricing tests how much developers are willing to pay for it.

DeepSeek said V4 Pro includes upgraded agent capabilities and adjustable reasoning effort, allowing developers to choose low, high, or maximum reasoning levels based on the task. The model supports OpenAI’s Responses API format and has been optimized for Codex workflows, lowering the friction for developers moving existing applications onto DeepSeek.

“We’re launching DeepSeek-V4-Pro today! 🚀V4 Pro is now available on app/web. Try it via “Expert Mode”.
V4 Pro is also available via API. Model names remain unchanged,” DeepSeek said in a post on X.

DeepSeek-V4-Pro arrives with some eye-catching benchmark results. The model scored 87.9 on Terminal-Bench 2.1, edging past Anthropic’s Opus 4.8 and coming close to Fable 5. It posted 83.3 on CyberGym and 62.7 on DeepSWE, according to benchmark figures released alongside the model.

V4 Pro is now available through DeepSeek’s app, website, and API, and has already been integrated into developer platforms including OpenCode.

The pricing shift comes at a pivotal moment for DeepSeek itself. TechStartups reported in July that the company was seeking a 500 billion yuan, roughly $74 billion, valuation in a new fundraising round as it prepared for a possible mainland China stock market debut, according to Bloomberg. The discussions followed DeepSeek’s first external financing round.

That fundraising push points to the bigger issue hanging over the AI industry. Training frontier models gets most of the attention, but serving those models to millions of users and developers can create enormous recurring infrastructure costs.

DeepSeek helped ignite an AI price war by proving that highly capable models could be offered at prices far below those of major U.S. competitors. Its latest move doesn’t end that strategy, but it does suggest the economics are getting harder to ignore.

V4 Pro may show how far DeepSeek’s models have come. The new API prices could reveal something equally interesting: **how much developers believe that performance is actually worth.**